Navigating a house requires a level of understanding that we still can't get using a top end GPU and optimal sensors.
"It is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility
This is a great example of "short term improvements are overestimated, long-term underestimated". As someone who jumped from a top-of-the-line robot vacuum in 2014 (not a Roomba - they were already slightly behind) to a high-quality-but-average one this year - those incremental improvements have added up to almost an entirely different product. That's not even counting the experiments out there like SwitchBot's plans (looks like the humidifier is out, but I don't see the dehumidifier on their site).
How has this space not seen more action on the consumer side? Roombas came out 20 years ago then there's been basically nothing but tiny increments on the same idea since.
Seems to me like an area that could make use the sensor/3d tech from self driving, but wouldn't need to be anywhere near as reliable to still be a useful product. You could probably charge Apple margins if you cracked it first. I'd pay $5-10k for something that would last a few years and could passively clean a bathroom or kitchen to a reasonable degree.